Computational analysis of genetic interaction network structures and gene properties

Abstract

Cellular systems are responsible for many complex tasks, such as carrying out cell cycle phases, responding to intra- and extra-cellular conditions, and resolving errors. Through analysis of biological networks, researchers have begun to describe how cells coordinate these processes by means of modularity and between-process connections. However, descriptions of this network-based cellular organization often do not incorporate the diverse characteristics and individual behaviors of the genes that compose it. Knowledge of gene properties and their relationships with biological network evolution is crucial for a complete understanding of cellular function, and investigation in this area can lead to general principles of biology that apply to many species. This dissertation will describe analyses of the Saccharomyces cerevisiae (baker’s yeast) genetic interaction network that connect gene topological behavior with various physical, functional, and evolutionary properties of genes. Genetic interactions occur between paired genes whose simultaneous mutations produce unexpected double-mutant phenotypes, which are indicative of a range of functional relationships. Because genetic interactions can be identified genome-wide in high-throughput experiments, their networks are comprehensive and unbiased representations of function to which we can apply computational methods that search for structure-function relationships. We begin by exploring the association between a set of gene properties and gene genetic interaction (GI) degree. Here, we build a decision tree model that sorts genes based on a set of properties, each of which has a correlation with GI degree, and accurately predicts GI degree. We show that our model, trained on S. cerevisiae, is also accurate for a very distant yeast species, Schizosaccharomyces pombe, demonstrating that the rules governing gene connectivity are well conserved. Finally, we used predictions from the model to identify gene modules that differ between the two yeast species. Next, we further characterize hub genes through an investigation of pleiotropy, the phenomenon of a single genetic locus with multiple phenotypic effects. Pleiotropy has typically been described by counting organism-level phenotypes, but a characterization based on genetic interactions can capture details about cellular processes that are buffered by the cell and never manifest in single mutant cellular phenotypes. For this analysis, we use frequent item set mining to discover GI modules, which we annotate with high-level processes, and use entropy to measure the functional diversity of each gene’s set of containing modules, thus distinguishing between genes whose functional influence is limited to very few bioprocesses and those whose roles are important for varied cellular functions. We identified a number of gene and protein characteristics that differed between genes with high and low pleiotropy and discuss the implications of these results regarding the nature and evolution of pleiotropy.